Analytical tools for the physicochemical profiling of drug candidates to predict absorption/distribution

Article

Reference

Analytical tools for the physicochemical profiling of drug candidates to predict absorption/distribution

HENCHOZ, Yveline, et al.

Abstract

The measurement of physicochemical properties at an early phase of drug discovery and development is crucial to reduce attrition rates due to poor biopharmaceutical properties.

Among these properties, ionization, lipophilicity, solubility and permeability are mandatory to predict the pharmacokinetic behavior of NCEs (new chemical entities). Due to the high number of NCEs, the analytical tools used to measure these properties are automated and progressively adapted to high-throughput technologies. The present review is dedicated to experimental methods applied in the early drug discovery process for the determination of solubility, ionization constants, lipophilicity and permeability of small molecules. The principles and experimental conditions of the different methods are described, and important enhancements in terms of throughput are highlighted.

HENCHOZ, Yveline, et al . Analytical tools for the physicochemical profiling of drug candidates to predict absorption/distribution. Analytical and Bioanalytical Chemistry , 2009, vol. 394, no. 3, p. 707-729

DOI : 10.1007/s00216-009-2634-y PMID : 19184676

Available at:

http://archive-ouverte.unige.ch/unige:1631

Disclaimer: layout of this document may differ from the published version.

1 / 1

REVIEW

Analytical tools for the physicochemical profiling of drug candidates to predict absorption/distribution

Yveline Henchoz&Bruno Bard&Davy Guillarme&

Pierre-Alain Carrupt_&Jean-Luc Veuthey_&

Sophie Martel

Received: 2 December 2008 / Revised: 16 January 2009 / Accepted: 19 January 2009 / Published online: 2 February 2009

#Springer-Verlag 2009

Abstract The measurement of physicochemical properties at an early phase of drug discovery and development is crucial to reduce attrition rates due to poor biopharmaceu- tical properties. Among these properties, ionization, lip- ophilicity, solubility and permeability are mandatory to predict the pharmacokinetic behavior of NCEs (new chemical entities). Due to the high number of NCEs, the analytical tools used to measure these properties are automated and progres- sively adapted to high-throughput technologies. The present review is dedicated to experimental methods applied in the early drug discovery process for the determination of solubility, ionization constants, lipophilicity and permeability of small molecules. The principles and experimental con- ditions of the different methods are described, and important enhancements in terms of throughput are highlighted.

Keywords Physicochemical profiling . Solubility . Ionization constant . Lipophilicity . Permeability . Experimental models

Abbreviations ACN Acetonitrile

ADME Absorption, distribution, metabolism and elimination

ADMET Absorption, distribution, metabolism, elimination and toxicity

APCI Atmospheric pressure chemical ionization API Atmospheric pressure ionization

APPI Atmospheric pressure photoionization b Gradient steepness

BBB Blood–brain barrier CAD Corona aerosol discharge

CAPS 3-(Cyclohexylamino)-1-propanesulfonic acid CCD Contactless conductivity detection

CE Capillary electrophoresis CNS Central nervous system

CNS+ Compounds crossing the blood–brain barrier CNS− Compounds not transported into the brain CZE Capillary zone electrophoresis

DAD Diode array detector DME Dimethyl ether DMSO Dimethyl sulfoxide

ELSD Evaporating light scattering detector EOF Electroosmotic flow

ESI Electrospray ionization

FaSSIF Fasted state simulated intestinal fluid FeSSIF Fed state simulated intestinal fluid HEPES 4-(2-Hydroxyethyl)piperazine-1-

propanesulfonic acid

HILIC Hydrophilic interaction liquid chromatography HPLC High-performance liquid chromatography HTS High-throughput screening

IAM Immobilized artificial membrane

ITIES Interface between two immiscible electrolyte solutions

k Retention factor

ki Retention factor at the initial gradient composition

LC Liquid chromatography

Leff Effective length of the capillary

LEKC Liposome electrokinetic chromatography DOI 10.1007/s00216-009-2634-y

Y. Henchoz

:

:

:

:

J.-L. Veuthey

:

School of Pharmaceutical Sciences, University of Geneva, University of Lausanne,

Quai E-Ansermet 30, 1211 Geneva 4, Switzerland e-mail: [email protected]

logD^pH Logarithm of the distribution coefficient at a given pH

logkIAM Logarithm of the retention factor measured on immobilized artificial membrane columns logkw Logarithm of the retention factor extrapolated

to 100% water

logP Logarithm of the partition coefficient logPa Apparent permeability coefficient logPe Effective permeability coefficient

logP^I Logarithm of the partition coefficient of an ion logPlip Logarithm of the partition coefficient in

liposomes/water system

logPoct Logarithm of the partition coefficient in the 1-octanol/water system

LOQ Limit of quantification

LSER Linear solvation free-energy relationship LSS Linear solvent strength

Ltot Total length of the capillary LYSA Lyophilized solubility assay

MEEKC Microemulsion electrokinetic chromatography MEKC Micellar electrokinetic chromatography MeOH Methanol

MES 2-Morpholinoethanesulfonic acid MS Mass spectrometry

NCE New chemical entity NMR Nuclear magnetic resonance NQAD Nano-quantity analyte detector ODS Octadecyl-bonded silica

PACE Pressure-assisted capillary electrophoresis PAMPA Parallel artificial membrane permeability assay PC Phosphatidylcholine

PC Polycarbonate

PDMAC Poly(diallyldimethylammonium chloride) PK Pharmacokinetic

pKa Ionization constant

pK_a^app Apparent ionization constant PVDF Polyvinylidene fluoride PVS Poly(vinylsulfonate) RI Refractive index

RPLC Reversed-phase liquid chromatography SAR Structure–activity relationship

SDS Sodium dodecyl sulfate t0 Column dead time tD System dwell time

tEOF Migration time of the electroosmotic flow marker

tG Gradient time tm Migration time

tmc Migration time of the micelle marker tEOF Migration time of the EOF marker tR Retention time of a compound

U Tension

UHPLC Ultrahigh-pressure liquid chromatography VEKC Vesicle electrokinetic chromatography

z Charge

Δϕ Change in composition of the mobile phase during the gradient

μeff Effective mobility

Introduction

Drug research is a complex and time-consuming process. A period of 12–20 years is needed from the discovery of new chemical entities (NCEs) to the marketing authorization of a new drug. This process can be divided into two main phases (cf. Fig.1): (i)drug discovery, which includes target identification, hit discovery, and lead optimization; and (ii) drug development, which comprises preclinical and clinical studies. Inappropriate pharmacokinetic (PK) behavior has been recognized as one of the major factors leading to the rejection of NCEs during drug development [1]. Therefore, a great deal of effort has been made to perform these studies as early as possible in the drug discovery process, preferably before the drug candidate enters the lead optimization phase.

In this way, only compounds with high potency and suitable PK properties are selected for development [2].

Physicochemical properties (solubility, ionization and lipophilicity) allow the ADME (absorption, distribution, metabolism and excretion) behavior of a compound to be predicted [3,4]. Solubility is a parameter of prime impor- tance in the drug discovery process. Indeed, drugs must be soluble in order to reach their targets, and poor solubility seems to be the most important cause of rejection during drug discovery and development [5]. Moreover, it severely affects results of in vitro screening bioassays.

Also, the majority of NCEs are ionizable compounds. As the ionization state of a compound affects its solubility and lipophilicity, and thus its pharmacokinetic behavior, early evaluation of its ionization constants is an important step in the discovery process [6]. Lipophilicity, which describes the partitioning of a compound between an aqueous and a lipidic environment, is another essential physicochemical property impacting ADME behavior, as it widely contrib- utes to membrane permeation, solubility, protein binding and metabolism.

Finally, important information on advanced ADME or PK properties such as passive drug permeability across a phospholipidic membrane is also useful. Previously, perme- ability experiments were generally carried out later in the discovery process, since they were based on low-throughput in vitro methods, often performed with biological materials.

Recently, new analytical methods based on artificial mem-

branes or partitioning in an anisotropic medium have emerged. These approaches are now largely implemented in the screening processes used in pharmaceutical research as a first predictor of biological membrane permeability.

Since advances in combinatorial chemistry have drastically increased the number of NCEs, tools for analytical physico- chemical and PK properties determination are automated and progressively adapted to high-throughput technologies [6].

Indeed, although in silico prediction methods are gaining in importance, experimental determination remains neces- sary in order to optimize in silico methods and build chemical libraries of experimental physicochemical data.

The present review is dedicated to experimental methods applied early in the drug discovery process to determine solubilities, ionization constants, lipophilicities and perme- abilities of small molecules. The principles and experimen- tal conditions of the different methods are described, and the important enhancements in terms of throughput are highlighted.

Ionization constants

Traditional methods

Potentiometric and spectrophotometric methods are com- monly used for pKadetermination.

In potentiometric titration (cf. Table 1), the pH of a vigorously stirred solution is continuously measured with a glass electrode as precisely known volumes of a standard- ized strong acid or base are added [7]. The pKa is determined from the difference between the potentiometric titration curve of the tested compound and a blank aqueous titration. When the solubility of a compound in aqueous

media is insufficient, co-solvents are added, and titrations at different co-solvent concentrations allow the pKa to be extrapolated to zero percent co-solvent [8–10]. This method is universal and takes 30–60 min/titration (i.e., 10–30 titrations/24 h) using an automated instrument (e.g., the Sirius GLpKa from Sirius Analytical Instruments, Forest Row, UK). As this technique requires milligram amounts of sample at relatively high concentrations (200–5000 μM) [11], a microscale pH-titrimetric method was described but is rarely used [12]. However, it remains difficult to handle impure or unstable compounds.

Spectrophotometric pKa determination (cf. Table 1) is based on the change in the absorption of a chromophore near ionizable moieties [4]. This technique was recently used for the pKa determination of lansoprazole [13] or sartans [14]. The ionization constants are determined from a plot of the absorbance measured at an appropriate wavelength as a function of pH. More recently, Tam and coworkers developed a generalized method called multiwavelength spectrophotometric titration. It requires a diode array detector coupled to an automated pH titrator and a much more complex data treatment [15–19]. Spectrophotometric pKa

determination is usually more sensitive than potentiometry, with sample concentrations ranging from 10 to 50 μM. A microscale spectrophotometric method was also introduced by Morgan et al. [12]. As it takes 30 min/titration, a spectral gradient analysis (SGA) method designed for a 96-well format and based on a pH gradient flow technique with diode array UV detection was developed to increase the throughput [20, 21]. A commercial SGA instrument (from Sirius Analytical Instruments) is available, allowing the determination of pKavalues of one compound in only 4 min (240 samples/24 h) [4]. However, this technique requires a measurable UV chromophore that is dependent on the pH, which is not the case for all pharmaceutical compounds [7, DRUG DISCOVERY DRUG DEVELOPMENT

Post-Launch activities

(Phase IV clinical trials)

In silico filters

LAUNCH

Detailed physico-chemical profiling

Solubility Ionization Lipophilicity Permeability In vitro HTS

Solubility Ionization Lipophilicity

Fig. 1 Scheme of the drug research process

11]. Moreover, it is also difficult to deal with impure and unstable compounds.

Capillary electrophoresis Theory and principles

In recent years, capillary zone electrophoresis (CZE) (cf.

Table 1) has emerged as the method of choice for pKa

determination due to some inherent advantages, as reported in several reviews [4,22–29]. (i) The sample and solvent consumptions are small: 1–50 nl of sample at concen- trations of 10–500 μM is usually introduced into the capillary, while the whole capillary volume is about 1μl (for a conventional capillary of length 50 cm and internal diameter 50 μm). (ii) Separations are performed under high electric fields with a flat flow profile, ensuring very thin peaks and high resolving power. (iii) The instru- mentation is fully automated and able to handle impure samples, as it is a separation method. (iv) This technique is universal, since different detection systems can be coupled to CZE [24].

With this approach, pKa determination is based on the measurement of the effective mobility (μeff) of the analyte at various pH values. Indeed, the μeff of a compound depends on the molar fraction of each ionized form present at a given pH. Practically speaking, μeff can be measured as:

m_eff ¼Leff Ltot

U 1

t_m 1 t_EOF

ð1Þ where tm and tEOF are the migration times of the analyte and neutral marker, respectively, U is the applied voltage (V), Ltot is the total capillary length (cm) and Leff is the effective capillary length (cm). The variation in electroos-

motic flow (EOF) at different pH values is corrected for by using a neutral marker, as described in Fig.2A.

Then, pKavalues are calculated by nonlinear regression using the relationship between μeff and pH (Eq. 2). This equation can be applied to different ionizable compounds, depending on the number and nature of the ionizable groups, as described elsewhere [30]:

m_eff ¼Xⁿ

i¼0

Qⁱ

j¼1

10^pKaj

" #

10^ðinÞpH Pⁿ

i¼0

Qⁱ

j¼110^pKaj

" #

10^ðinÞpH

m_HniX^zi ð2Þ

where iand j are, respectively, theith and jth dissociation steps,nis the total number of ionizable groups andzin the charge of the fully protonated species HnX^z. A typical plot of the variation of the effective mobility of the monobasic compound lidocaine as a function of pH is presented in Fig. 2B.

The determination of pKa values by CZE has mainly been applied to pharmaceutical compounds [28]. Agree- ment with other methods is generally good (about 0.2 pKa

units), but can be much lower in the case of weak bases (pKa<3) or weak acids (pKa>10).

Experimental conditions

To obtain reliable pKa values by CZE, several general considerations concerning critical experimental conditions, like the choice of the buffer (i.e., its nature and ionic strength), the applied voltage and the detection method, must be taken into account.

First, the buffer nature should be carefully chosen.

Rafols et al. published an extended study of this problem [31]. Ammonium buffers should be avoided due to their Table 1 Experimental techniques for pKadetermination

Method Potentiometry Spectrophotometry CZE-UV

Measurement Ionization profile Ionization profile Single points

Quantity 2–10 mg 2–10 mg <<<1 mg

High purity Necessary Necessary Not necessary

Low/medium-throughput instrumentation

Automatic titrator: GLpKa (Sirius Analytical Instruments, Forest Row, UK)

DAD-UV spectrophotometer coupled to an automatic titrator (various vendors)

Capillary electrophoresis unit (various vendors)

High-throughput instrumentation

with 96-well plate technology

- Spectral gradient analysis, SGA

(Sirius Analytical Instruments, Forest Row, UK)

Multiplexed capillary electrophoresis:

cePRO 9600 (Advanced Analytical Technologies, Ames, IA, USA) or MCE 2000 (Pfizer Laboratory, Groton, CT, USA)

Miscellaneous - Only for compounds with a

measurable chromophore that is dependent on the pH

-

instability, as they are highly volatile. Amino-based buffers like butylammonium, ethanolammonium and diethanolam- monium tend to interact with the capillary wall, generating distortions in the curves of μeff versus pH. Some other buffers interact with certain types of analytes, which could affect their migration behavior [30–35]. For example, it is well known that borate buffers interact strongly with compounds possessing consecutive diols, like catechol.

Finally, interactions between mono/dihydrogenophosphate, MES, HEPES and protonated amines may occur when the pH is higher than the pKa, and similarly between borate buffers and deprotonated amines when the pH is lower than

the buffer pKa. Therefore, taking into account the con- straints discussed above, recommended buffers have been extracted from the literature for the whole pH range, and are listed in Table 2. Another important parameter concerning buffers is their ionic strength. A constant ionic strength should be used to avoid μeff variations and keep activity coefficients constant. If each buffer has a different ionic strength, corrections for ionic strength must be calculated for each μeff measurement. Since the different formulae used forμeffcorrections are approximate [36,37], a decrease in the accuracy of the results must be expected [35]. Therefore, it is easier to work at constant ionic strength and then correct the pKa values determined with activity coefficients for zero ionic strength. The ionic strength is generally set to between 10 and 50 mM to ensure sufficient buffer capacity and low Joule heating.

Buffers cover a large pH range (e.g., from 2 to 12), generally with an increment of 1.0 pH unit. For compounds with multiple pKavalues close to one another, more buffers should be used (for example with an increment of 0.5 pH units) to obtain good curve fitting [38].

Temperature variations should be strictly avoided when determining pKa values, since this parameter is dependent on the temperature [39]. Numerous parameters influence the temperature inside the capillary, such as the applied voltage, the buffer conductivity, the capillary dimensions (diameter, total length) and the thermoregulation system (controlled by an air- or liquid-pooling system). Therefore, the applied voltage should be adjusted to remain in the linear region of Ohm’s law and prevent excessive Joule heating [40]. In the absence of active cooling of the capillary, it is possible to correct the measured mobilities to standard temperature with empirical equations [41].

Organic or hydroorganic solvents can alternatively be used to determine dissociation constants of water-insoluble

Table 2 Recommended buffers according to [30]

pH Buffer pKa Possible interactions

1.5–2.5 Phosphate 2.12

3.0–4.5 Formate 3.75

4.0–5.5 Acetate 4.76

5.5–7.0 MES^a 6.13 Protonated amines

6.5–8.0 MOPS^b 7.20 Protonated amines 6.5–8.0 Phosphate 7.20 Protonated amines 7.5–9.0 Tricine^c 8.15 Protonated amines 8.5–10.0 CHES^d 9.30 Protonated amines

8.5–10.0 Borate 9.24 Diols

11.5–12.0 Phosphate 12.37 Protonated amines

aMES: 2-morpholinoethanesulfonic acid

bMOPS: 3-(N-morpholino)propanesulfonic acid

cTricine: N-tris(hydroxymethyl)methylglycine

dCHES: 2-(N-cyclohexylamino)ethanesulfonic acid

1 3 5 7 9 11 13

0.0 0.5 1.0 1.5 2.0 2.5

pH µeff [10-4 cm2 V-1 s-1 ]

pK_a=7.80 min

0.5 1.0 1.5 2.0

mAU

0 10

mAU

0 10

0.5 1.0 1.5min mAU

0 10

min

0.5 1.5

lidocaine EOF

lidocaineEOF

EOF + lidocaine

pH 5.0

pH 7.5

pH 10.0

b a

Fig. 2 AElectropherograms of the monobasic compound lidocaine and the EOF marker at different pH values: at pH 5.0 the compound is mainly in its cationic form and migrates before the EOF, at pH 7.5 the compound is partially ionized and migrates closer to the EOF, and at pH 10.0 the compound is neutral and migrates with the EOF.BThe curve of the effective mobility of lidocaine as a function of pH is sigmoidal: at pH<6 the compound is mainly present in its cationic form and μeff is high, between pH 6 and 10 the curve shows an inflection point corresponding to the pKaof lidocaine, and at pH>10 the compound is neutral andμeffis equal to zero

or sparingly soluble compounds. Methanol and acetonitrile are the organic solvents most commonly used in CZE, with a preference for methanol as it has properties closer to those of water. Numerous papers deal with pKadetermination by CZE in aqueous–organic solvents [33,42–56] or nonaqueous solvents [57–62]. Practically speaking, the pH of hydro- organic buffers is measured with electrodes calibrated with the usual aqueous standards, which leads to the absolute

s

wpH scale (i.e., pH measured in solvent s, relating to the water pH scale) [63]. Then the aqueous pKa values (i.e.,

wwpKa) can be estimated from the^s_wpKavalues determined in aqueous–organic solvents by Yasuda–Shedlovsky extrapola- tion [64]. The pH of an organic buffer is measured with electrodes calibrated with organic standards prepared by mixing equimolar quantities of acids and their conjugates with known^s_spK_avalues. This leads to the relative^s_spH scale (i.e., pH measured in solvent s, relating to the solvent pH scale). The^w_wpKa values can be estimated from nonaqueous solvents by means of linear extrapolation equations derived for similar types of compounds [41].

The UV detector is the preferred one for pKadetermina- tion by CZE, as it is sensitive enough for the majority of pharmaceutical compounds. Moreover, it enables a simulta- neous spectrophotometric determination of pKavalues (i.e., from UV spectra measured at the maxima of the electro- phoresis peaks), as described by several authors [51, 65–

68]. The pKavalues of non-UV-absorbing compounds can be determined by indirect UV detection [69, 70], ampero- metric detection [71] or contactless conductivity detection (CCD) [30,36,37]. The latter is particularly interesting, as it can be easily implemented in CZE and may provide interesting complementary information to photometric detec- tion. Finally, mass spectrometry (MS) can be coupled to CZE [72]. This detector presents several advantages in terms of sensitivity and universality. Furthermore, several compounds can be analyzed simultaneously due to its high selectivity, increasing the throughput of pKa determination. However, the big challenge of pKadetermination by CZE-MS is to find volatile and stable enough buffers for the whole pH range.

Towards high-throughput methods

Numerous efforts have been recently made to increase the throughput of pKadetermination by CE. A major challenge with conventional CE methods is the long migration times at low pH due to the limited EOF. An interesting way to overcome this problem is to use dynamic coatings. For anions, a suitable strategy consists of using a positively charged coating, generally polybren, in negative mode [69, 73–75]. Neutral polyacrylamide coatings eliminating the EOF can be utilized for cations in positive mode and anions in negative mode [32, 43, 76]. Finally, the most generic strategy is to employ a double coating (polybren and poly

(vinylsulfonate) (PVS) or CEofix^®), thus creating a high EOF whatever the buffer pH [14,30,77].

The use of short capillaries is not recommended due to the high electric fields, which generate significant Joule heating.

On the other hand, short-end injection is a good strategy as it reduces the effective length of the capillary without enhancing Joule heating [14,30,55,72,78,79]. The latter strategy coupled with a dynamic coating procedure enabled pKadetermination in about 2 h per compound [30].

To increase the throughput, pressure-assisted capillary electrophoresis (PACE) was developed for pKadetermina- tion [67,79–84]. The application of an external pressure of up to 2 psi does not seem to affect pKadetermination but it does increase peak broadening. Therefore, it may decrease the precision of tm and tEOF determination, and μeff

measurement. The throughput of such a PACE method was reported to be about 20 compounds per day [81,82], but it can be further increased to 50 compounds in less than 3 h using MS detection due to sample pooling [72].

The throughput obtained by instruments with a single capillary is limited by the number of experimental points needed to fit the curves of μeff versus pH. More recently, several applications using vacuum-assisted multiplexed 96-channel capillary electrophoresis with UV detection have been reported for high-throughput pKascreening (24 compounds/h) [23,56,85–87]. The ability to adapt existing single-capillary electrophoresis methods to a multiplexed instrument proves that CE has its place in drug discovery and development.

Lipophilicity

Lipophilicity is expressed as the logarithm of the partition coefficient (logP) for partitioning between two immiscible solvent phases, and is valid for a single electrical species [88]. The distribution coefficient, expressed as log D^pH, refers to the weighted contributions of all electrical forms of an ionizable compound present at a given pH [22]. The 1-octanol/water system is the widely accepted reference system for the determination of lipophilicity.

Traditional methods

The classical shake-flask technique has little evolved over the decades but remains the reference method for lipophilicity measurement [89]. Briefly, the compound is mixed with two immiscible solvents, generally water and 1-octanol, until equilibrium is reached. Then the two phases are separated and the solute concentration is usually measured in both of them. However, this procedure is tedious, time-consuming and sensitive to impurities. It is also prone to emulsion problems and requires large amounts of sample. Therefore,

in the last decade, different strategies have been developed to speed up, automate and miniaturize the shake-flask approach. An automated shake-flask system for high-through- put log Dmeasurement (48 samples/day, in duplicate) was described by Hitzel et al. [90]. The entire liquid handling is performed by a robotic system, and the partitioning process takes place on a 96-well plate. After shaking, each phase is directly injected from the plate into the LC-UV system with an auto-injector. A fast generic gradient allows the concen- tration of the solute of interest to be measured. An alternative is to use an LC-MS system [91]. A commercially available instrument using an automated and miniaturized shake-flask method is manufactured by Analiza (Cleveland, OH, USA) [92, 93]. More recently, a high-speed log D (HSLogD) measurement system based on automated sampling was developed to reduce contaminations from the 1-octanol phase when sampling the water phase [94]. This contamination was prevented by aspirating a plug of water before sampling the water phase. Despite all of these improvements, this technique still requires some time-consuming steps that multiply the manipulations involved, like the mutual satura- tion and decantation of both phases. In addition, the measurable logPoctrange remains limited (from−3 to 4).

When dealing with ionizable compounds, dual phase- potentiometric titration is possible. This method is based on the pKa shift that occurs in the presence of a partitioning solvent, for instance 1-octanol. A large difference between the true and apparent pKavalues indicates a large logPvalue. A log P determination requires two titrations and directly provides a distribution profile rather than single points [22].

This method presents the same features as pKadetermination in terms of sample consumption and concentration, and the same automated instrument (Sirius GLpKa from Sirius Analytical Instruments) can be used. However, this tech- nique is only appropriate to ionizable compounds, and it is difficult to handle weak lipophilic bases with low pKa or weak acids with high pKavalues.

Cyclic voltammetry at the interface between two immis- cible electrolyte solutions (ITIES) is the method of choice for studying the logPof ions (logP^I) [22,95]. The distribution of a particular ion at the ITIES is determined by how similar its electrochemical potentials in the two phases are, and log P^Ican be deduced from the Nernst equation at the ITIES.

Capillary electrophoresis Theory and principles

As reported in several reviews [23, 25, 26, 29, 96–102], capillary electrophoresis is emerging as an interesting separative method for log Poct determination due to its inherent advantages, as enumerated in the relevant subsec-

tion of“Ionization constants”(also, cf. Table 3). For this Table3ExperimentaltechniquesforlogPdeterminationoct MethodShakeflaskPotentiometryCyclicvoltammetryMEEKC-UVLC-UV MeasurementDirectsinglepointsDirectdistributionprofileIndirectsinglepointsIndirectsinglepointsIndirectsinglepoints logPrange−3to40to8−8to0−1to7−1to8oct Quantity2–10mg2–10mg1–10mg<<<1mg<1mg HighpurityNecessaryNecessaryNecessaryNotnecessaryNotnecessary Low/medium-throughputShakingsystemandAutomatictitrator:GLpKa instrumentationdetector(SiriusAnalyticalInstruments, ForestRow,UK) Electrochemical systemCapillaryelectrophoresisunit (variousvendors)Liquidchromatographyunit (variousvendors) High-throughput instrumentation with96-wellplatetechnology

Automatedshake-flask systemADW(Analiza, Cleveland,OH,USA) --Multiplexedcapillaryelectrophoresis: cePRO9600(AdvancedAnalytical Technologies,Ames,IA,USA)or MCE2000(PfizerLaboratory, Groton,CT,USA) Ultrahigh-pressureliquid chromatography(various vendors) MiscellaneousTediousprocedureOnlyapplicabletoionizable compoundsOnlyforionicspeciesMicroemulsionstabilitymightbe anissueSecondaryinteractionscan happen

purpose, a CE mode that is able to separate neutral analytes (i.e., one combining the features of conventional CE and LC) is needed. Therefore, different CE modes (including a pseudo-stationary phase) have been tested, namely MEKC, MEEKC and VEKC/LEKC (micellar, microemulsion and vesicle/liposome electrokinetic chromatography, respectively).

MEKC is an electrokinetic chromatography technique that uses buffers containing micelles as a pseudo-stationary phase. The separation of neutral compounds is due to differential partitioning between aqueous and micellar phases.

Micellar pseudo-stationary phases, commonly sodium dodecyl sulfate (SDS), are considered to mimic biological membranes better than 1-octanol or RPLC stationary phases [25]. Practically speaking, logPoctdetermination by MEKC is based on the measurement of the retention factork:

k¼ t_Rt_EOF 1_t^t_MC^R

tEOF

ð3Þ

wheretR,tEOFandtMCare the migration times of the analyte, neutral marker and micelle marker, respectively. Highly hydrophilic neutral compounds such as acetone are used as EOF markers, whereas highly lipophilic compounds such as dodecaphenone are selected as micelle markers. Good linear correlations were found between logk(measured by MEKC) and logPoctfor a wide variety of compounds and conditions [103–122]. However, congeneric behavior was observed [108,116] and confirmed by LSER analyses, which demon- strated differences in H-bond basicity and dipolarity/polariz- ability for partitioning into SDS micelles or the 1-octanol/

water system [98].

LEKC and VEKC are also electrokinetic chromatography techniques that use liposomes or vesicles (i.e., bilayer structures enclosing an aqueous core region) as a pseudo- stationary phase [102]. Linear correlations between log k measured by LEKC/VEKC (Eq. 3) and log Poct were reported [123–126], and these techniques appear to mimic physiological membranes more closely than ME(E)KC [127]. However, vesicles and liposomes are unstable and difficult to prepare reproducibly. Therefore, their use remains limited for logPoctdetermination [99].

Another electrokinetic chromatography technique using buffers containing nanometer-sized oil droplets (MEEKC) has been widely used, as it is the most appropriate CE mode for log Poctdetermination. MEEKC separations are based on the same principles as MEKC, LEKC and VEKC.

Numerous linear correlations between log k measured by MEEKC (Eq.3) and logPoctwere reported in the literature [86, 128–145], and calibration curves with standard com- pounds generally allowed accurate log Poctdeterminations over a large logPoctrange (from−1 to 7) [133–136, 140, 142,144]. Indeed, microemulsion systems are known to be more stable and reliable than MEKC for logPoctestimation

[129], as they more closely resemble phospholipidic vesicles than SDS micellar systems do [128]. LSERs analyses also suggest that microemulsion systems are good models for the 1-octanol/water partitioning process [142, 146].

However, it should be noted that the investigated acidic or basic compounds must be in their neutral forms, since their ionized forms may generate ion pairs with surfactants and additional electrophoretic migration. This additional electro- phoretic migration can be corrected for when ion-pair interactions are not involved or are very weak [147].

Experimental conditions

The most critical experimental conditions for log Poct

measurement by MEEKC, like the composition of the microemulsion system, need to be pointed out.

Microemulsions are complex ternary mixtures of water, oil and surfactant. Therefore, the microemulsion buffers used in MEEKC are made up of many different compo- nents. Variations in the nature and concentrations of these components can affect the microemulsion stability and migration. Microemulsions are stabilized by a surfactant and a cosurfactant, which both reduce the interfacial tension between the oil–water interface. The cosurfactant molecules lie between the surfactant ones at the surface of the microemulsion, thus reducing the repulsion between the surfactant head groups. The most widely used surfactant, oil and cosurfactant in MEEKC are SDS (anionic), heptane and 1-butanol, respectively [99,148–151]. A typical micro- emulsion system used for logPoctdetermination consists of 6.5% w/w 1-butanol, 0.8% w/w heptane and 1.4% w/

w SDS. However, some authors have used concentrations of up to 3.3% w/w SDS, both to increase the micro- emulsion stability and to widen the separation window, especially for compounds with high logPoctvalues [142].

Although phosphate and phosphate/borate buffered systems have mainly been used for logPoctmeasurements, zwitterionic buffers such as CAPS are being used more and more due to their lower conductivities [142]. Indeed, even if temperature has less of an effect on log Poct than pKa

measurements, accurate and reproducible results should be obtained in a thermostated environment [89].

Most applications were performed with UV detection. To the best of our knowledge, MS detection has not been used to increase the throughput of log Poct determination by MEEKC. Indeed, microemulsion buffers contain nonvola- tile components such as SDS, which could generate ion suppression as well as ionization source contamination and clogging. Therefore, the sensitivity decreases dramatically when using ESI-MS detection [26]. To overcome this problem, it is now possible to switch from ESI (electro- spray ionization) to APPI (atmospheric pressure photoion- ization) in CE. Indeed, the latter offers a higher

compatibility with nonvolatile buffers and has a similar sensitivity to CE-ESI-MS [152–154].

Towards high-throughput methods

As for pKa determination by CZE, several efforts have recently been made to increase the throughput of logPoct

determination by MEEKC.

A major problem is the long migration time at low pH due to the limited EOF. Microemulsion systems usually contain an anionic surfactant (SDS) and are used under alkaline con- ditions. However, it is sometimes necessary to work at low pH where the EOF is limited or even nonexistent, for example when determining log Poct values of acidic compounds. To overcome this problem, the polarities of the electrodes can be reversed. However, separate experiments with positive polarity are needed to determine the EOF mobility. A more elegant solution to this problem is to use sulfonic acid-coated [132, 133] or dynamically coated capillaries with adsorbed bilayer structures such as poly(diallyldimethylammonium chloride) (PDMAC) and PVS [138] or polybren and PVS [139], which provides a sufficient EOF at low pH.

Other rapid approaches for logPoct determination have been developed. For instance, pressure-assisted MEEKC can improve the throughput to up to 48 samples per day using a single-capillary system [135], while an instrument and software dedicated to this experiment in an automated format with 96 parallel capillary channels allow high- throughput logPoctdeterminations (46 compounds/h) [86, 140, 142]. The latter is probably the most interesting strategy for drug discovery and development.

Liquid chromatography Theory and principles

The general methodology of lipophilicity determination by LC has been the subject of several reviews [89, 96–98, 100–102, 155–159]. Indeed, the indirect reversed-phase liquid chromatography (RPLC) approach is considered to be the best alternative to the direct shake-flask method for lipophilicity assessment, since it offers numerous advan- tages (cf. Table3). (i) The sample consumption is small:

100μl of sample at a concentration of 10–500μM. (ii) The instrumentation is fully automated, compatible with the 96- well plate format, and the sample throughput is high. (iii) There is no requirement for highly pure samples, as it is a separation method. (iv) This technique is universal, since different detection systems can be coupled to LC. (v) The measurable logPoctrange is large (from −1 to 8).

The ability of RPLC techniques to assess lipophilicity relies on the similarity between the retention mechanism on RPLC stationary phases and octanol–water partitioning.

The determination of log Poct by RPLC is based on the measurement of the retention factor k of the investigated compound between the mobile and stationary phase:

log k¼log tR

t₀ 1

ð4Þ wheretRis the retention time of the investigated compound and t0is the column dead time. The extrapolation of logk values to 100% water, which leads to log kw values, is a convenient means of standardizing chromatographic lip- ophilicity parameters.

Lipophilicity measurements can be performed in either isocratic or gradient mode. Under isocratic conditions, logk values are calculated from tR measurements at various concentrations of the organic solvent in the mobile phase (Eq. 4), as described in Fig. 3A. When using methanol, isocratic logkvalues are linearly correlated with the organic modifier percentage ϕ in the mobile phase (as exhibited by Fig. 3B):

log k¼log k_wSϕ ð5Þ

where S is a constant that depends on the nature of the compound and the organic modifier. With acetonitrile, the correlation between log k and the organic modifier percentage in the mobile phase is quadratic [160]:

log k¼log kwþBϕþAϕ² ð6Þ

where A and B are constant for a given compound and organic modifier.

Numerous linear correlations between log Poct and log kw obtained in isocratic mode have been reported in the literature for different classes of compounds [161,162], as also shown in Fig. 3C. Valko et al. also observed linear relationships between logPoctandϕ0, which correspond to the organic modifier concentration (from 0 to 100%) that produces an equimolar distribution between the stationary and mobile phases (i.e., when logk=0) [163,164]. Finally, it should be noted that several linear correlations between log Poct and isocratic log k values at different organic modifier percentages have been described [165–173].

However, all these linear relationships are generally better with structurally related compounds (cf., congeneric behav- ior) or a low number of analytes.

Gradient approaches were proposed to speed up lip- ophilicity determination. They are generally based on the linear solvent strength (LSS) theory elaborated by Snyder and Dolan, where the analyte retention time (tR) in a linear- gradient separation can be expressed as follows [174]:

tR¼t0

b log 2:3ð kibþ1Þ þt0þtD ð7Þ wherekiis thekvalue at the initial isocratic composition of the gradient,t0is the column dead time (min), andtDis the

system dwell time for gradient elution (min). The gradient steepness parameter b can be described by the following relationship:

b¼t0ΔϕS

t_G ð8Þ

wheretGis the gradient time from the beginning to the end of the gradient (min), andΔϕis the change in composition

during the gradient (from 0 to 1). The two unknown parameterskiand S in Eqs. 7 and 8 can be obtained from two gradient runs differing only in gradient time. These two values then allow estimation log kw directly with Eq. 5 when using methanol or Eq. 6 when using acetonitrile.

Appropriate calculation procedures for log kw and S are included in available modeling software, such as Drylab (Rheodyne, Rohnert Park, CA, USA) or Osiris (Datalys, Grenoble, France) [175–177]. An alternative strategy which does not require complex calculation procedure involves assuming that S is constant for structurally similar com- pounds. Although this approach is evidently less accurate than the former, it allows estimation log kw from a single gradient run [92, 178–182]. Based on this assumption, Valko and coworkers introduced the chromatographic hydrophobicity index (CHI), derived from the ϕ0 scale [183–188]. Numerous linear relationships between a reten- tion parameter obtained in gradient mode and log Poct or log D values have been reported in the literature. In conclusion, it can be stated that the gradient mode is a more generic and faster strategy than the isocratic mode.

However, data treatment remains complex and it is not as accurate as the isocratic mode [175,189,190].

Experimental conditions

The choice of several experimental conditions, such as the organic modifier and the stationary phase, is crucial for log Poctdetermination by RPLC.

Mobile phase Methanol and acetonitrile are the two organic modifiers most commonly employed for log Poctdetermi- nation. In the literature, methanol was generally preferred to acetonitrile [89,101, 160]. Indeed, quadratic extrapolation (Eq. 6) is less accurate than linear extrapolation (Eq. 5) [191]. Moreover, as a protic solvent, methanol can interact with free silanol groups, reducing secondary interactions between analytes and silanol groups. Additionally, during equilibration, MeOH molecules form a monolayer at the surface of the stationary phase, which provides a hydrogen- bonding capacity in better agreement with 1-octanol/water partition [157].

In the last few years, many authors have shown that the addition of a small amount of 1-octanol to the methanol fraction of the mobile phase (usually 0.25% V/V) and the use of 1-octanol-saturated buffers improves the correlation between logPoctor logDand logkwvalues, especially in the case of basic compounds [157, 192–201]. Octanol seems to form a coating on the silica support of the column during equilibration, conferring octanol-like properties to the stationary phase. Additionally, it may act as a weak masking agent, reducing secondary interactions caused by 0.00

0.50 1.00 1.50

0.00 0.50 1.00

AU ^70%

60%

80%

50%

40%

0 1 2 3 4 5

0 1 2 3 4 5

log P_oct log kw

0 20 40 60 80 100

-0.5 0.0 0.5 1.0 1.5 2.0 2.5

% MeOH (V/V)

log k

min

log k

=1.96

log P

=2.15 log k

=1.96

a

b

c

Fig. 3 AChromatograms of propiophenone at different percentages of methanol.BPlot of logkobtained for propiophenone as a function of the percentage of methanol, and extrapolation of the logkwvalue to 100% water by linear regression. C Calibration line allowing the calculation of the logPoctvalue of propiophenone from the obtained logkwvalue

residual silanol groups [201]. However, the precise mech- anism through which 1-octanol influences the retention is still not fully established. It should also be noted that the addition of 1-octanol to the mobile phase is a tedious and time-consuming strategy due to the saturation of buffers and the long equilibration times, which are crucial for the repeatability of the measurements.

Stationary phasesOctadecyl-bonded silica (ODS) station- ary phases have been used extensively for lipophilicity measurements for over 30 years. However, silanophilic interactions due to the free silanol groups (including hydrogen bonding as well as electrostatic interactions) interfere in the partitioning mechanism of these columns.

To overcome this problem, apolar and polar end-capped stationary phases were developed. However, the latter had no real advantage over classical ODS columns for log Poct

determination [202]. On the contrary, polar embedded stationary phases with an amide, carbamate, ether or sulfonamide group are considered better than endcapped ones for lipophilicity estimation due to the electrostatic shielding provided by the polar group, thus reducing the access to the free silanol groups [202]. However, other interactions between polar embedded groups and analytes may occur. More recently, high pH-resistant stationary phases allowing logPoct determination of highly basic compounds have been launched, such as bidentate stationary phases and organic/inorganic hybrid stationary phases [201]. Hybrid stationary phases with embedded polar groups are the best choice for lipophilicity determination by RPLC [202]. Using such columns, there is no need to add a masking agent such as triethylamine and n-decylamine because of the lower amount of free silanols. Finally, monolithic silica stationary phases were also tested for fast lipophilicity determination, but their use remains limited [182].

Immobilized artificial membrane (IAM) columns [156, 158,203] were widely investigated for lipophilicity deter- mination. Briefly, IAM columns contain different types of phospholipidic monolayers covalently bonded to silica particles and are generally less retentive than ODS columns due to their polar head groups. They allow direct measure- ment of log kw values for compounds with moderate lipophilicity, while up to 30% acetonitrile is used in the case of highly lipophilic compounds. There is not always a clear relationship between retention properties measured by IAM chromatography (logkIAMwor logkIAM) and logPoct

values. For neutral and acidic compounds, logkIAMwvalues are generally linearly correlated with logPoctvalues despite a less hydrophobic environment. However, there is a systematic deviation in the case of basic and zwitterionic compounds. Indeed, the retention mechanism on IAM columns is based not only on lipophilic interactions but

also on specific ionic interactions, which are particularly important with positively charged compounds that interact with the phosphate anions of the stationary phase [200].

Consequently, retention on IAM stationary phases is only moderately affected by the presence of positive charges.

The distribution coefficients measured in a 1-octanol/water system may thus underestimate the potential of positively charged compounds to interact with biomembranes [204–

207]. Practically speaking, IAM chromatography raises several issues, since its retention properties are dependent on the experimental conditions and a column-aging phenomenon was reported by several authors [156,208].

Recently, a new approach using hydrophilic interaction chromatography (HILIC) has been applied to lipophilicity determination. This separation mode is based on polar stationary phases (underivatized silica or silica bonded with polar groups like diols, amide, cyanopropyl, aminopropyl) with partly aqueous eluents containing a relatively hydro- phobic miscible solvent, generally acetonitrile [209–212].

The retention mechanism is considered to be a complex mix of partitioning and adsorption of the compounds between the eluent and a water-rich layer partially immobi- lized at the surface of the polar stationary phase. There are a number of different interactions depending on the mobile and stationary phases involved: hydrophilic retention when high proportions of organic modifiers are used in the mobile phase, and a reversed-phase-type retention when the mobile phase contains high proportions of water. An additional ion exchange mechanism may occur on ionic HILIC stationary phases [213,214]. Bard et al. measured retention factors of highly basic compounds in their cationic forms on a ZIC- pHILIC stationary phase based on a grafted polymeric layer with sulfoalkylbetaine zwitterioinic moieties on a wide-pore silica. As lipophilicity expresses the difference between hydrophobicity and polarity, the difference between the isocratic logkvalues measured at 0% and 95% acetonitrile (Δlogk0-95) was linearly correlated to their logPoctvalues (r²=0.93) [213]. Therefore, this method offers a new and promising way to measure log Poct values of highly basic compounds without working under extreme pH conditions.

From a practical point of view, it should be noted that it is necessary to strictly control the nature of the buffer, the pH and the ionic strength, as the retention mechanism implies ion exchange (ionic competition).

Detection methods UV detection is commonly used for log Poct determination by LC. However, some alternative detectors have also been used to deal with UV-inactive compounds. For instance, refractive index (RI) and evap- orating light scattering detectors (ELSD) were used in several studies because of their quasi-universality [165, 171,213]. In the last few years, several other aerosol-based detectors have been launched. Corona aerosol discharge

(CAD) and nano-quantity analyte detectors (NQAD), which are also able to deal with nonchromophoric compounds, are potentially more interesting than RI and ELSD due to their higher sensitivity. Finally, if the cost per analysis is not an issue, the detector of choice is MS, due to its high selectivity and quasi-universality. Indeed, the different atmospheric pressure ionization (API) sources available (electrospray ionization, atmospheric pressure chemical ionization and atmospheric pressure photoionization, ESI, APCI and APPI, respectively) allow a wide variety of compounds to be dealt with [168,181,185,190,215–218].

Towards high-throughput strategies

Several strategies were applied to speed up lipophilicity determination by LC, such as the use of short columns and high flow-rates. Both approaches decrease the analysis time and extend the range of measurable logPoct. However, they cause a significant reduction in the chromatographic performance and the accuracy of the results [179,180,184,193,195].

Recently, the development of columns packed with small porous particles (sub-2μm) has enabled a substantial improvement of chromatographic performance in LC.

Indeed, this technology enables a significant time reduction without compromising efficiency. However, such columns generate high backpressures (<1000 bar) and require dedicated instrumentation [219–221]. An ultrahigh-pressure liquid chromatography (UHPLC) system was used for log Poct measurements, offering a significant increase in the throughput (by at least a factor of 12 compared to HPLC) [190,222]. Moreover, due to the stability of Acquity BEH Shield RP18 hybrid columns at high pH (up to 11), logPoct

values of basic compounds with pKa values ranging from 6.3 to 10.3 could be determined. However, these basic compounds displayed a different behavior from the optimal set of calibration compounds selected by cluster analysis.

Nevertheless, accurate logPoctvalues of basic compounds could be obtained in isocratic mode with 1-octanol addition or when using isocratic logkvalues at 50% methanol [190,222].

The latter strategy is very interesting in terms of average log Poct determination time, since a single run is required (ca.

5 min per compound). Finally, MS coupling allowed the throughput to be further increased (by an additional factor of 10) due to sample pooling. It should be noted that precise knowledge of the column dead time, extra-column volume and delay time is mandatory and extremely critical in UHPLC configuration for accurate retention factor determination.

Simultaneous determination of pKaand logPoct

Kaliszan et al. developed methods allowing the simulta- neous determination of pKaand logPoctvalues from a set

of experiments carried out on the same HPLC instrument.

Apparent pKaand logPoctvalues could be determined in 3–

4 gradient runs. Two organic modifier gradient runs allowed log Poctvalues to be determined (see the“Theory and principles”subsection of the“Liquid chromatography”

section) and the organic modifier percentage to be estimated, providing an appropriate retention coefficient for the nonionized analyte. Then, one pH-gradient run carried out with the previously determined eluent compo- sition and one isocratic run to get the retention factor of the ionized form allowed the apparent pKa values to be determined [175, 176, 189, 223–225]. The obtained log kw values correlated well with log Poct values. However, there was only a poor agreement with literature pKavalues.

Therefore, this generic approach was further improved and a second method requiring 9 or 18 pH/organic modifier double-gradient runs was evaluated. Its success was limited, as the correlation with logPoctvalues was lower than that seen for the previous method, and there was only a moderate agreement with literature pKavalues [226,227].

Moreover, this strategy is time-consuming (9–18 runs) and requires complex data treatment.

Solubility

Solubility measurements are fundamental in drug absorp- tion, as insufficient solubility can interfere with both pharmacokinetic and pharmacodynamic properties.

Two solubility values can be measured depending on the experimental method used, namely thermodynamic or kinetic solubility. Solubility assays in discovery and development must be adapted to phase properties and requirements (Table 4). In drug development, high-quality solubility data are needed by chemists to facilitate appro- priate decisions to be made so as to overcome potential solubility liabilities. Additionally, solid-state properties (purity, degree of crystallinity, particle size and polymor- phism) of the tested compounds are studied and character- ized in detail. Therefore, an evaluation of thermodynamic solubility is required and is often considered the gold standard. Nevertheless, it is becoming increasingly com- mon to measure it earlier in the discovery process.

Solubility measurements are also performed in biorelevant media to identify in vitro/in vivo correlations in accordance with the Biopharmaceutical Classification System (BCS) [228, 229]. This classification has been developed to identify the fundamental rate-limiting biopharmaceutical factors of intestinal drug absorption, and classifies drug compounds into four classes based on their solubility and intestinal permeability (Fig.4). This system typically takes aqueous solubilities measured in buffer media across the whole pH range 1–7.5 as input.

In drug discovery, kinetic solubility is the most relevant parameter for rapidly evaluating whether the compound, which is generally predissolved in dimethyl sulfoxide (DMSO), stays in solution after dilution in specific screening media. Incomplete solubility could indeed lead to unreliable results, underestimated activity, reduced HTS hit rates and an inaccurate structure–activity relationship (SAR) when not identified [230]. For these reasons,

experimental conditions are adapted to the conditions of the bioassay.

Reference methods Thermodynamic solubility

The shake-flask method is considered the method of reference for solubility measurement. An excess of solid sample is added to the medium in a flask, and the resulting suspension is shaken until equilibrium solubility (i.e., thermodynamic solubility) is reached, generally after 24 h.

Undissolved sample is then removed by filtration or centrifugation, and the concentration of the compound in the filtrate is determined. Effects of crystal lattice and polymorphic forms are considered in this process, which is important, as they can significantly influence the solubility [231]. Therefore, thermodynamic solubility is often consid- ered the“true”solubility of a compound. Nevertheless, this method suffers from several weaknesses: (i) relatively large amounts of compounds (several mg) are required to perform the measurements; (ii) the individual weighing step is time-consuming and difficult to automate [232]; and (iii) long incubation times are required to reach equilibrium solubility. Therefore, this method is mainly used in advanced studies of lead compounds, which are often Table 4 Differences between the methods used for solubility measurement in drug discovery and drug development processes

Discovery Development

Compounds tested

Number 100–1000 10

Quantity available A few mg > g

Purity Limited Improved

Solid state Amorphous or partially crystalline (not characterized)

Stable, crystalline material (characterized) Distribution Generally in DMSO stock solutions Generally in solid form

Methods

Type of solubility measured Kinetic solubility (fully dependent on experimental conditions)

Thermodynamic solubility

Throughput High Low 25–50 compounds a week

Automation Fully Only partially automated

Format 96-well or 384-well microplates Small scale (single tube)

Incubation time Minutes Hours or days

Detection UV, turbidity HPLC-UV, HPLC-MS

Media Aqueous >20 (aqueous, organic, biorelevant media,

formulations, excipients…) Data generated and intended

purpose

Solubility in screening bioassay media to avoid misinterpretation

Solubility and dissolution in biorelevant media

Rank-order hits Evaluation of formulations

Characterization and optimization of solid state Selection of promising compounds

Development of adequate strategies to overcome solubility problems

Solubility

Permeability

Class I

High permeability High solubility

Class II

High permeability Low solubility

Class III

Low permeability High solubility

Low permeability Low solubility

Class IV

Fig. 4 Schematic representation of the Biopharmaceutical Classifica- tion System